@jenz-ai/claude-agent-sdk

Zero-config Jenz observability for the Claude Agent SDK. One import-line change captures every query() call as an observable Run with model calls, tool invocations, token usage (including cache reads/writes), and subagent activity.

Install

npm install @jenz-ai/claude-agent-sdk @anthropic-ai/claude-agent-sdk

Usage

// before
import { query } from '@anthropic-ai/claude-agent-sdk';

// after — one line changed
import { query } from '@jenz-ai/claude-agent-sdk';

for await (const msg of query({ prompt: 'fix the bug' })) {
  // your iteration unchanged
}

Set JENZ_API_KEY in your environment. Get a key at https://jenz.dev/api-keys.

What gets captured

| Source | Captured as | |---|---| | query() call starts | A new Run (framework=claude-agent, agentType=claude_code, agentName=options.agent ?? 'claude-agent') | | SDKAssistantMessage | An llm_call event with provider=anthropic, model, input/output/cache tokens | | PreToolUse / PostToolUse hook | A tool_call event correlated by tool_use_id | | PostToolUseFailure | A failed tool_call event | | SubagentStart / SubagentStop | A tool_call event named subagent:<agent_type> | | SDKResultMessage (success) | Run.finish({status:'completed', output, metadata:{costUsd, durationMs, numTurns, ttftMs}}) | | SDKResultMessage (error_*) | Run.finish({status:'errored', errorMessage: subtype}) |

Advanced: access the active Run

import { query, getActiveRun } from '@jenz-ai/claude-agent-sdk';

for await (const msg of query({ prompt: 'go' })) {
  const run = getActiveRun();
  if (run?.signal.aborted) break; // honor dashboard "Stop"
  // ...
}

Dormant by default

If JENZ_API_KEY is unset, the adapter prints one warning per process and passes every query() straight through. Your agent runs untouched.

Custom backend

Set JENZ_BASE_URL to point at a self-hosted Jenz API.

Documentation

Full docs at https://jenz.dev/docs.